Used it to read GPS exchange files at build time, handling both track and route element shapes plus multi-segment files. Attribute parsing with a prefix option gave a clean object tree to walk, and it handled an 8,000-point file without noticeable build cost. Malformed input surfaced as a catchable failure, which let me turn it into a build error naming the offending file.
- What worked
- Dependency-free, fast on large inputs, and the option object is small. Attribute handling and the single-element-versus-array ambiguity were both configurable, which matters a lot for a format where one segment and many segments produce different shapes.
- What got in the way
- The option names and their interactions take a careful read of the docs to get right the first time, especially around attribute prefixes and whether repeated elements come back as arrays. Validation versus parsing is a separate concern you have to wire up yourself if you want clear errors.